Generation landslide susceptibility maps are to study the relations among the image data of band variables
concerning the occurrence/nonoccurrence on investigated samples of landslide. A feasible solution on generating
landslide susceptibility map through land cover classification of remote sensing data is an important topic for studies of image processing and classification. Image classification considering clustering technique is well-accepted when ground truth data is scarce. However, applying the clustering technique, the initial guess of cluster centers may lead to different results. As a result, the traditional K-means data clustering technique may fail to arrange the data to the appropriate target groups. Accordingly, this study employs bacterial foraging algorithm (BFA), which successfully resolves the image data of clustering problems in landslide. On the other hand, the constrained clustering is a useful clustering technique to improve the classification outcomes when few label data are available. Accordingly, the study focused on the classifier by using BFA optimized constrained clustering to study landslide area in which the evaluation of landslide occurrence by remote sensing image data is rationally studied. The results show constrained BFA clustering
yields the better classification results (93.7%) than those of BFA clustering (81%) and K-means (77.6%).
Keywords: Landslide, image classification, bacterial foraging algorithm